-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathSTING.py
203 lines (149 loc) · 7.86 KB
/
STING.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
#This code has been partly copied from https://github.com/JinmiaoChenLab/GraphST since STING uses GraphST as the outer GNN
import torch
from .preprocess import preprocess_adj, preprocess, construct_interaction, add_contrastive_label, get_feature, permutation, fix_seed, get_gene_graph, get_gene_graph_cluster
import time
import random
import numpy as np
from .model import Encoder
from tqdm import tqdm
from torch import nn
import torch.nn.functional as F
#from scipy.sparse.csc import csc_matrix
#from scipy.sparse.csr import csr_matrix
from torch_geometric.data import Data
from sklearn.decomposition import PCA
from sklearn.cluster import KMeans
import scanpy as sc
class STING():
def __init__(self,
adata,
device= torch.device('cpu'),
learning_rate=0.001,
weight_decay=0.00,
epochs=900,
#dim_input=3000,
dim_output=64,
random_seed = 2,
alpha = 10,
beta = 1
):
'''\
Parameters
----------
adata : anndata
AnnData object of spatial data.
device : string, optional
Using GPU or CPU? The default is 'cpu'.
learning_rate : float, optional
Learning rate for ST representation learning. The default is 0.001.
learning_rate_sc : float, optional
Learning rate for scRNA representation learning. The default is 0.01.
weight_decay : float, optional
Weight factor to control the influence of weight parameters. The default is 0.00.
epochs : int, optional
Epoch for model training. The default is 600.
dim_input : int, optional
Dimension of input feature. The default is 3000.
dim_output : int, optional
Dimension of output representation. The default is 64.
random_seed : int, optional
Random seed to fix model initialization. The default is 41.
alpha : float, optional
Weight factor to control the influence of reconstruction loss in representation learning.
The default is 10.
beta : float, optional
Weight factor to control the influence of contrastive loss in representation learning.
The default is 1.
Returns
-------
The AnnData object including the new embeddings and attention scores.
'''
self.adata = adata.copy()
self.device = device
self.learning_rate=learning_rate
self.weight_decay=weight_decay
self.epochs=epochs
self.random_seed = random_seed
self.alpha = alpha
self.beta = beta
fix_seed(self.random_seed)
adata1 = adata.copy()
sc.pp.highly_variable_genes(adata1, flavor="seurat_v3", n_top_genes=3000)
depth = np.sum(adata1[:, adata1.var['highly_variable']].X)
del adata1
dim_input = self.features.shape[1]
if 'highly_variable' not in adata.var.keys():
preprocess(self.adata, dim_input)
if 'adj' not in adata.obsm.keys():
construct_interaction(self.adata)
if 'label_CSL' not in adata.obsm.keys():
add_contrastive_label(self.adata)
if 'feat' not in adata.obsm.keys():
get_feature(self.adata)
km = KMeans(n_clusters=self.adata.obsm['spatial'].shape[0]//100, random_state=self.random_seed).fit_predict(self.adata.obsm['spatial'])
self.adata.obsm['km'] = km
self.km_graphs = []
num_genes = self.adata.obsm['feat'].shape[1]
spars = np.sum(self.adata.obsm['feat'] == 0)/self.adata.obsm['feat'].size
avg_neigh = min(7, ((depth/100000)**0.5) * (0.35*(np.log(num_genes-30)-1)))
for i in range(np.max(km) + 1):
km_g = get_gene_graph_cluster(self.adata.obsm['feat'][km == i], avg_neigh = avg_neigh) #km_a
self.km_graphs.append(km_g.to(self.device))
self.features = torch.FloatTensor(self.adata.obsm['feat'].copy()).to(self.device)
self.features_a = torch.FloatTensor(self.adata.obsm['feat_a'].copy()).to(self.device)
self.label_CSL = torch.FloatTensor(self.adata.obsm['label_CSL']).to(self.device)
self.adj = self.adata.obsm['adj']
self.graph_neigh = torch.FloatTensor(self.adata.obsm['graph_neigh'].copy() + np.eye(self.adj.shape[0])).to(self.device)
if dim_output > self.adata.obsm['feat'].shape[1]:
dim_output = self.adata.obsm['feat'].shape[1]
pca = PCA(n_components=dim_output, random_state=42)
pca_features = pca.fit_transform(self.adata.obsm['feat'].copy())
self.inner_graph_features = []
for i in range(self.features.shape[0]):
gene_exp_i = self.features[i]
gene_exp_i = gene_exp_i.reshape(-1,1)
gene_edges = self.km_graphs[km[i]]
inner_data = Data(x=gene_exp_i.float().to(self.device), edge_index=gene_edges)#, edge_attr = edge_weights)
self.inner_graph_features.append(inner_data)
self.inner_graph_features_a = []
for i in range(self.features_a.shape[0]):
gene_exp_i = self.features_a[i]
gene_exp_i = gene_exp_i.reshape(-1,1)
gene_edges = self.km_graphs[km[i]]
inner_data = Data(x=gene_exp_i.float().to(self.device), edge_index=gene_edges)#, edge_attr = edge_weights)
self.inner_graph_features_a.append(inner_data)
self.dim_input = self.features.shape[1]
self.dim_output = dim_output
self.adj = preprocess_adj(self.adj)
self.adj = torch.FloatTensor(self.adj).to(self.device)
def train(self):
self.model = Encoder(self.dim_input, self.dim_output, self.graph_neigh, self.device).to(self.device)
self.loss_CSL = nn.BCEWithLogitsLoss()
self.optimizer = torch.optim.Adam(self.model.parameters(), self.learning_rate,
weight_decay=self.weight_decay)
print('Begin to train ST data...')
self.model.train()
for epoch in tqdm(range(self.epochs)):
self.model.train()
self.features_a = permutation(self.features)
self.inner_graph_features_a = []
for i in range(self.features_a.shape[0]):
gene_exp_i = self.features_a[i]
gene_exp_i = gene_exp_i.reshape(-1,1)
gene_edges = self.km_graphs[self.adata.obsm['km'][i]]
inner_data = Data(x=gene_exp_i.float().to(self.device), edge_index=gene_edges)
self.inner_graph_features_a.append(inner_data)
self.hiden_feat, self.emb, ret, ret_a, _, _, _ = self.model(self.features, self.features_a, self.adj, self.inner_graph_features,self.inner_graph_features_a)
self.loss_sl_1 = self.loss_CSL(ret, self.label_CSL)
self.loss_sl_2 = self.loss_CSL(ret_a, self.label_CSL)
self.loss_feat = F.mse_loss(self.features, self.emb)
loss = self.alpha*self.loss_feat + self.beta*(self.loss_sl_1 + self.loss_sl_2)
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
print("Optimization finished for ST data!")
with torch.no_grad():
self.model.eval()
self.adata.obsm['emb'] = self.model(self.features, self.features_a, self.adj, self.inner_graph_features,self.inner_graph_features_a)[0].detach().cpu().numpy()
self.adata.uns['a1'] = self.model(self.features, self.features_a, self.adj, self.inner_graph_features,self.inner_graph_features_a)[-2]
return self.adata